Skip to main content

Storage FAQs

What should I do if the system disk space is insufficient?​

Most reasons for insufficient instance space are because directories such as /root/, /usr/local/miniconda3, /tmp, /opt occupy too much space, and users placed large files such as datasets or data on the system disk instead of the /gz-data data disk. The /root/ directory usually stores caches for pip and conda, and /usr/local/miniconda3 stores virtual environments by default. Too many virtual environments or packages in one environment can cause the /usr/local/miniconda3 directory ...

You can view the root directory disk usage with the following command. If the system disk shows full, use the following commands to locate which directory is consuming space, then move or clean it.

# 1. Check instance system disk usage
df -h | grep "/$" | awk '{print "System disk usage: "$5"\nTotal space: "$2"\nUsed space: "$3"\nAvailable space: "$4}'

# 2. List and sort disk usage of all files under / directory
du -h --max-depth=1 --exclude=/proc --exclude=/gz-data --exclude=/gz-fs / | head -n -1 | sort -hr
# Example output
14G     /usr
180M    /root
92M     /tmp
50M     /var
1.9M    /etc
...

# 3. Further analyze which subdirectory consumes the most space, e.g., /usr
du -h --max-depth=1 /usr | head -n -1 | sort -hr
# Example output
9.7G    /usr/local
3.2G    /usr/lib
130M    /usr/share
109M    /usr/bin
31M     /usr/include

# 4. Further analyze /usr/local
du -h --max-depth=1 /usr/local | head -n -1 | sort -hr
# Example output
5.9G    /usr/local/miniconda3   
3.9G    /usr/local/cuda-11.7   # CUDA installation path in instance system. Do not delete this directory. If CUDA is removed, CUDA cannot be used in the instance. This belongs to the system image layer and is not counted in the system disk’s 30G capacity.
16M     /usr/local/bin
4.0K    /usr/local/sbin
tip

The /usr/local/miniconda3 directory, which installs conda and virtual environments, is often the main cause of system disk overuse. If this directory is too large, you can use the conda clone function to clone an environment from /usr/local/miniconda3 to /gz-data/, then delete the original environment in /usr/local/miniconda3 to free system disk space.

Example steps:

# List current virtual environments
conda info -e 

# Clone an environment (GPUGEEK) to /gz-data/gm-env
conda create -p /gz-data/gm-env --clone GPUGEEK

# Activate new environment and verify
conda activate /gz-data/gm-env 

# After confirming it works, remove old environment to free space
conda remove -n GPUGEEK --all

What should I do if the data disk space is insufficient?​

# Check instance data disk usage
df -h | grep "/gz-data$" | awk '{print "Data disk usage: "$5"\nTotal space: "$2"\nUsed space: "$3"\nAvailable space: "$4}'

If available space is low, go to the GPUGEEK Console, find the corresponding instance, click More → Resize Data Disk, enter the new capacity, then click Confirm to expand the data disk.